On some level, Eli Lilly’s partnership with computing giant Nvidia feels inevitable. The first pharma to reach a $1 trillion valuation teaming up with a tech firm that has ridden the AI wave to a record-setting $5 trillion valuation of its own—a Wall Street supergroup, a la Cream, Audioslave or boygenius.
But Lilly leaders told Fierce Biotech that a baser instinct is driving the drug giant’s decision to invest big in Nvidia and AI: anxiety.
“We’re trying to escape the traditional pharma industry life cycle,” Diogo Rau, Lilly’s chief information and digital officer, told Fierce. “This industry is so bizarre compared to other industries, with these huge peaks and these really deep troughs that span decades.”
With Lilly currently climbing to perhaps the highest peak in pharma history thanks to the success of its diabetes and weight loss drug tirzepitide, Rau wants to avoid the comedown.
“What can you do to break out of that cycle?” Rau asked. “It’s probably the thing that I’m the most concerned about, for the long term.”
It’s here where Nvidia—and the AI applications its computing power enables—come in.
On Wednesday, Lilly leaders, including Rau and Chief AI Officer Thomas Fuchs, Ph.D., cut the ribbon on a new Nvidia supercomputer, named LillyPod, that the Big Pharma believes to be the most powerful in the industry.
After first announcing the supercomputer in October, Lilly and Nvidia followed up with a $1 billion commitment for a new Bay Area AI co-innovation lab, announced at the January J.P. Morgan Healthcare Conference in San Francisco.
The goals of syncing up with Nvidia are very broad, Rau told Fierce, with the aim of bringing together Lilly’s top scientists and Nvidia’s top model builders and seeing what happens.
“It’s really a beautiful combination of very orthogonal capabilities and interests,” Fuchs agreed in an interview with Fierce. “Nvidia is not going to be a medicines company and Lilly will not start producing our own GPUs (graphics processing units).”
“For now,” Fuchs added with a smile, with Rau noting that the pharma’s contract with Nvidia doesn’t prevent Lilly from someday doing so.
Related
Lilly’s growing bond with Nvidia fits with the company’s overall strategy, outlined by oncology head Jake Van Naarden for Fierce last December, to become “a backbone of the overall innovation ecosystem all around the world.”
Other Lilly efforts along these lines include TuneLab, a program where companies can access models trained on Lilly’s data in return for contributing more data of their own, and incubators called Gateway Labs housed in biotech hubs like San Francisco, Boston, San Diego, Philadelphia, Beijing and Shanghai.
The pharma also collaborates intensely with academic institutions like Indiana University, Purdue University, Massachusetts Institute of Technology and the California Institute of Technology, Fuchs said.
“We really try to be the tip of the spear [of] what you can do in AI, in medicine making, and really float all boats,” he said.
Fact and Fiction
To say that AI has been grossly overhyped, including in the life sciences, would be a severe understatement at this point. On this front, Rau has no qualms about offering a reality check.
“The hype is actually a serious threat to the research itself,” the tech leader said. “Because if the hype becomes the story, then we’re all going to be disappointed.”
One major area that’s been overblown is AI’s claimed ability to speed up drug development, a famously slow process.
“There’s a tendency to think that we’re now going to be able to discover new medicines in three months,” Rau said. “That’s one that’s particularly damaging and destructive.”
To be sure, LillyPod will help the Indianapolis drugmaker speed up development timelines, Rau added, but in piecemeal ways. By automating clinical trial tasks like patient enrollment and optimizing manufacturing processes, he hopes Lilly can cut the typical 10-year timeline for a new drug down to five years.
Manufacturing is one area where Rau thinks the hype is real, he said, and Lilly uses AI “all over the place” in its production process. For example, an auto-injector—like those used for weight loss blockbuster Zepbound—can be photographed 70 or 80 times in a “split second” and analyzed for defects by AI.
“We’ve also used AI extensively in forecasting,” Rau added, which has greatly improved the drugmaker’s supply and demand balance.
When it comes to creating new medicines, though, the power of LillyPod is difficult to fathom. The supercomputer is built from 1,016 GPUs, each one of which is more powerful than the Cray-2 supercomputer that Lilly bought in 1989.
A single LillyPod GPU is “7 million times as powerful” as Cray-2, Rau explained. Put 1,000 of them together, and it’s “7 billion times as powerful as a supercomputer that was critical to research” back in the 1990s.
This much power is needed to cope with the incredible scale of biology’s complexity, Chief AI Officer Fuchs said.
“We need hundreds of GPUs to just find one antibody, one new molecule that would bind to a target, because the sheer complexity of a single cell goes far beyond what our human language can even describe,” Fuchs said. “That already shows why you need that compute [power] in discovery and in molecular chemistry and in biology, and that’s why it’s also so exciting.”
Related
Lilly plans to turn LillyPod toward building AI models using a foundation of the pharma’s many decades of research, Fuchs said, some of which has never seen the light of day.
“People think if you ingest all of [scientific] literature, you could train a co-scientist or an artificial scientist,” he said. “That does not work, because that would only be the experiments that were successful.”
For better or worse, Lilly has reams of data on experiments that have flopped during its 150-year history. With all that information fed into different models that can specialize in making different kinds of medicines, and with Lilly’s budding interest in automated labs, Rau thinks there’s little a human scientist can do that AI someday can’t.
“The speed of progression of these models, it does make me question how long we’re going to be able to beat the machines,” he said.
But there is one area where humans will always win out over AI, Fuchs chimed in: curiosity.
“AI in its form today, these are fabulous tools, but it’s still a piece of software,” Fuchs said. “It doesn’t have will or volition.”
“The human curiosity to actually drive your army of [AI] agents into one direction or another direction, to actually come up with something new, that’s fundamentally human,” he added.